Recetly package lists were expanded by options and they now have their
own configuration file. This makes them incompatible with configuration
done by bootstrap script. This updates bootstrap script to allow usage
of package list options and also removes invalid generation of lists to
updater's config.

The configuration is not replaced by any easy way because we are not
using it and generating another configuration is just too big of a
hustle. Instead we should investigate in future some other way to
generate configuration (for example using package) instead of this.

This fixes medkit generation. It was failing because new version of
updater-ng now uses autotools instead of custom makefile.

Also usign is no longer required for medkit generation (by updater).
Updater now has its own implementation using OpenSSL. It is still
required for signing.

This adds data collection plugin to datacollect package list.

There is an updated kernel which fixes CVE-2020-10757 and contains
fixes for SDIO Wi-Fi card (mwifiex driver), which means there shouldn't be memory corruption so far.
To make things even worse, there is an updated wireguard.

This fixes restarts of hostapd process by netifd when wpad (hostapd)
program file is removed/replaced.

This happes on update and tiggers start of new hostapd instance while
old one is still running.

Makefile was modified in upstream. This is just simple rebase of
existing patch.

This package is necessary to control rescue modes for Turris MOX.

Karel Koci authored 4 years ago and Karel Koci committed 4 years ago

This is going to be superseded by LinkSignature in packages.

This implementation was problematic in the way that it touched version
of package after it was used in other places. The result were invalid
dependencies between some packages.

Karel Koci authored 4 years ago and Karel Koci committed 4 years ago

This adds new pakage list called drivers. It provides users with
alternative core drivers.
Currently user can choose ath10k-ct drivers and choose between standard
or htt firmware.

Drivers do not explicitly collide but they provide same files so new
file with such collisions was created and this conflict was noted in it.

This solves issue that tos3to4-early depends on switch-branch and that
one depends on updater-supervisor and that one depends on updater-ng.
In the end that creates with additional dependency of updater-ng on
tos3to4-early cycle dependency and updater won't allow that with
critical package.

This just uses tos3to4-early for same purpose as previously updater-ng.
It is just rude fix to break cycle.

This should in future replace potentially automatic package version
bumping patch. This should be cleaner solution for same problem as it
won't tamper with versions and does not need server-side database.

Problem with tampering with version is that matching version exactly was
then pretty much broken. Once package was bumped the version did not
match. It is impossible to update dependency easily enough when package
bump occurs.

The idea is that updater just matches signature that should be
consistent between builds (unless dynamic link dependencies changed of
course). When build results to package with different link dependencies
then signature changes and updater reinstall such package to version
from repository that is consistent. Version stays the same all the time.

Signature is generated from two sources. Dynamic library names and
dynamic symbols that are linked to those libraries are included.

This patch was backported by Kernel developers to LTS versions.
It can be found in 5.4.48, 4.4.185 and 4.4.228.

We are secure minded and sending metrics unencrypted is not in our plan.

This sets kernel identity to packaging@turris.cz. It should cause
initial line in kernel log to contain this instead of build server
user and host name.

Fixes compilation problem with node package feed
https://github.com/nxhack/openwrt-node-packages/issues/679

We updated all our build servers to Debian Buster, but LuCI is not
compiled. It ends up with following error:
make[2]: execvp: /usr/bin/env: Argument list too long

I found that there are patches on mailing list of OpenWrt-devel [1].
They were not reviewed neither applied since February. Let's apply them
to see if it it works.

[1] https://patchwork.ozlabs.org/project/openwrt/list/?series=159823

On Turris 1.x it seems that some assember code is not correct and result
are failures for some domain's certificates such as google.com or
pypi.org.

This reverts commit 444fcd15.

Co-authored-by: Vojtech Myslivec <vojtech.myslivec@nic.cz>